Next Army is a collaborative series by CSIS Futures Lab and the Modern War Institute launched in honor of the U.S. Army’s 250th birthday and the Army Transformation Initiative (ATI). The commentaries explore how emerging technologies, organizational reforms, and major shifts in the strategic environment will shape the force of 2040 and beyond.
In the future, the U.S. Army will operate as a distributed, data-centric network in which every echelon—from squad to theater armies and corps serving as combined forces land component commands—can tap a continuously refined “brain of the Army.” This agentic AI model will integrate lessons harvested from professional military education (PME) and real-world operations, fusing human insight with machine speed to accelerate the Military Decision-Making Process (MDMP), Troop Leading Procedures (TLP), and functional planning. Commanders will query staffs that combine human military professionals and an ecosystem of AI agents fine-tuned through reinforcement learning from human feedback. The best and brightest across the Army will collectively teach AI agents how to think about land power based on insights captured in classrooms, staff rides, and decision games—turning the classroom and leader development into the structured data needed to train algorithms.
Widely used AI platforms ingest commercial datasets divorced from the realities of land warfare, leaving commanders with generic models that misunderstand terrain, tempo, and tactical nuance. Most foundation models, which are generalists, are trained on more data about the Kardasian family than corps commanders from World War II or evolution of Army doctrine for leading large units.
As a result, the Army of the future must find mechanisms for fine-tuning foundation models that will become ubiquitous across the force. The ideal location to this end already exists: military education. Transforming PME into a more dynamic setting and deliberate data-aggregator capturing outlines for courses of action, commander’s intent statements, and after-action reflections creates the type of human-based, high-context data needed for broader AI adoption in the profession of arms. This change will require that every schoolhouse become more of a battle lab in the best tradition of the profession. There will be little time for lectures on strategy and outdated history treatments as students focus on fighting each other and even AI agents replicating threat doctrine. Civilian academics will still play a key role, but all curriculum—from history to political science and international relations—will be calibrated to the core purpose: creating a battle lab where thinking leaders fight and data on how they think is captured along the way. This vision is not cost-neutral. The Army will need to invest in the computational infrastructure and the processes required to harvest data from this battle lab.
Reimagining the General Staff in an Era of Agentic AI
There is a long history of thinking about how best to hone judgment in the profession of arms. In 1890, Spencer Wilkinson published The Brain of an Army, a popular account of the rise of the German general staff and its key role in planning successful military campaigns in the late nineteenth century. Successive generations of military professionals used the Prussian model and ideas about education to map out how to educate “enlightened soldiers.” At the core, this process involved a mix of campaign analysis through historical cases and decision games, captured in Carl von Clausewitz’s On War in the idea of critical analysis in Book 2, Chapter 5. The true professional had to balance studying the past—to accumulate insights about the complexity of warfare—with an eye toward the future battles they might wage. In other words, a synthesis between data (history) and larger ideas about how to fight (theory).
Today, that synthesis is still the heart of teaching military professionals how to visualize, describe, and direct major operations. Yet, the “data” does not reside in historical cases alone. Rather, it is more organic and generated each time a military professional solves a problem or analyzes a course of action. Even classroom debates, if captured, become a source of data bout the character of war and ideas for gaining an advantage. To the extent the schoolhouse aligns its curriculum and assessments against the objective of winning future wars, this data is also measurable. The Army can harvest the best insights based on evaluation, thus ensuring the data used to train models reflects the upper half and not the lower half of the profession.
In this context, reinforcement learning from human feedback (RLHF) becomes a modern extension of the nineteenth-century ideal of cultivating judgment through structured problem-solving. In an Army schoolhouse designed to train AI agents, RLHF would involve observing how students—human military professionals—engage with complex planning problems, wargames, and operational dilemmas. Their responses, feedback, and corrections could be systematically captured and transformed into preference datasets that help train AI agents to emulate high-quality military reasoning. The best student-generated courses of action, decision rationales, and critiques would serve as reward signals for tuning agents designed to support future staff processes.
Over time, the schoolhouse becomes not just a place for human learning, but also a dynamic training environment for AI agents, where doctrine, planning tools, and decision support systems evolve alongside the officers who use them. Just as the Prussians institutionalized staff rides and campaign studies to train their general staff, the Army can now institutionalize RLHF pipelines to build an “AI-enabled brain of the Army”—one that learns from its best practitioners to support the many.
Transforming Army Schoolhouses into RLHF Engines
As the Army looks to the future and deeper AI integration, it should start with harvesting data from across its schoolhouses. This will require three policy changes.
First, the secretary of the Army and the Army’s chief of staff should mandate that PME capture data and structure it in a manner that lends itself to support fine-tuning. The Army should issue a policy directive requiring all schoolhouses—from Captain’s Career Courses to Senior Service Colleges—to incorporate RLHF data capture into their curriculum and assessments. This includes embedding structured decisionmaking exercises, planning artifacts, and red-team critiques that feed directly into machine learning pipelines. To enable this transformation, the Army must invest in modern classroom infrastructure—cloud-native platforms, secure annotation environments, and AI-integrated learning management systems. It will also require augmenting faculty with technical staff who understand both military planning and data science. The outcome: Every educational environment becomes a dual-use platform that trains both people and AI agents for future war.
Second, the Army will need a lead agency to manage the project and should designate U.S. Army Training and Doctrine Command (TRADOC) for the task. TRADOC should serve as the Army’s lead agent for collecting, curating, and analyzing PME-generated data to support RLHF pipelines. This includes developing doctrinally aligned data-tagging schemas—covering elements like mission analysis frameworks, operational variables, principles of war, theories of advantage, and tactics—and establishing consistent benchmarks to validate model fine-tuning. TRADOC must also oversee accuracy and relevance using techniques like expert review panels, adversarial testing, and comparative analytics across schoolhouses. To scale this mission, the Army should fund partnerships with external research organizations, such as the National Institute of Standards and Technology and AI labs affiliated with universities and think tanks, ensuring model development aligns with responsible AI standards and national-level safety benchmarks.
Third, incentives matter. The Army will need to reward human feedback as a new form of professional contribution. The time spent in the schoolhouse and simulating modern battles should be rewarded, with top performers gaining recognition. The Army can incentivize participation by offering soldiers micro-credentials for contributing high-quality feedback to RLHF loops. Officers could earn credit for validating AI-generated staff estimates, refining doctrinal heuristics, and surfacing edge-case scenarios missed by machines. This approach turns every PME seminar into a dual-reinforcement loop—one that improves officer judgment and simultaneously trains AI systems tuned to the operational art. To do so requires creating the right mix of incentives and adapting an outdated evaluation system.
Educating the Army and Its Algorithms
Building the “brain of the Army” will not happen by accident. It requires a deliberate redesign of professional military education as both a learning environment and a data-generating engine. This is not just about adopting AI. It is about teaching AI how the Army thinks, using the very institutions designed to cultivate judgment, intuition, and initiative in human leaders.
The Prussians created the general staff to institutionalize their approach to planning and thinking about war. The U.S. Army now has an opportunity to create something just as revolutionary: a living, learning warfighting network where PME fuels an iterative loop between human and machine, where every wargame, course of action, and critique becomes training data, and where the best ideas don’t vanish into lecture notes but live on as part of an evolving agentic AI ecosystem. The payoff is an Army whose decision aids anticipate intent, whose planning tools learn from its most gifted leaders, and whose digital infrastructure adapts faster than the threat.
If the Army is serious about transformation, it must turn its schoolhouses into battle labs, its students into contributors to a collective military mind, and its education system into the scaffolding of the next revolution in military affairs. That means investing not just in technology, but in people, infrastructure, and purpose.
Benjamin Jensen is director of the Futures Lab and a senior fellow for the Defense and Security Department at the Center for Strategic and International Studies in Washington, D.C.
